Understanding space plasma properties via laboratory generation and AI/ML approaches

Abstract

This thesis presents the fabrication and characterization of a hermetically sealed plasma tube operating at 1 nano atmospheric pressure, exhibiting a plasma temperature of 14800.470 K and an electron number density of 4.90⇥1014, consistent with existing literature. These parameters were determined through spectroscopic analysis conducted using a UV-Optical-Infrared sensor module, capable of achieving a minimum resolution of 0.197nm with an integration time of 10 seconds. Building upon this experimental foundation, the thesis explores the modeling of space weather plasma phenomena through AI-assisted prediction of the Dst storm index using various machine learning and deep learning LSTM-based approaches. Initially, models were trained without considering ring current feedback, utilizing solely solar wind parameters. The Theil-Sen regression method yielded promising results, achieving a correlation coefficient of 0.6456, with a normalized standard deviation of 0.77, an RMSE of 9.72nT, and a value of 0.73. Subsequently, an Artificial Neural Network (ANN) trained without ring current feedback exhibited comparable performance, with a correlation coefficient of 0.67. Upon integrating ring current feedback from the previous hour into the models, significant enhancements in predictive accuracy were observed. The MLP Regression model, incorporating ring current feedback, outperformed other approaches, achieving an RMSE of 3.05nT and a normalized standard deviation of 0.24. Similarly, an ANN trained with ring current feedback exhibited superior performance, with a correlation coefficient of 0.97, an NSD of 0.23, and an RMSE of 3.28nT, as optimized through grid search.